Fig. 3

Scatter plot of (sample based) adjusted estimates of the treatment effect (b₁) as a function of the confounder imbalance between the two groups (Cohen’s d), with true parameter values β₁ = 1 and β₂ =  − 1, for continuous and binary confounders, at sample size 2 × 50. Grey lines represent non-parametric regression functions (Lowess)